The emergence of the World Wide Web represented a new advertising opportunity available to any company: The possibility of global exposure to a large audience at a very small cost. As a result, a whole new industry has emerged by offering services related to search advertising, in which an advertiser pays for a prominent position in lists of ads. In order to maintain the credibility and market share of the service that conveys them - for example, a search engine - such ads must be displayed only to users who are interested in them, on what is called Targeted Advertising. Therefore, those services need to use a recommendation system that can choose which ads show to which users. Recommendation systems based on collaborative filtering use the preferences of other users as features to a learning system, since such preferences can be quite detailed, generating recommendations not only for the most popular items but also to item niches. In this work, we develop an ads recommendation system that applies Collaborative Filtering based on matrix factorization to the problem of predicting the Click-Through Rate, a Targeted Advertising metric that expresses the relevance of a particular ad for the users searching for a specific keyword. In order to validate the proposed method of Click-Through Rate prediction, we carry out several experiments on a synthetic data set. Additionally, the work contributes to the design of LearnAds, a framework for ads recommendation systems based on Machine Learning.